Route Planning for Autonomous Vehicles

The present disclosure relates to automated driving technology, and more particularly, to a system and method that improves route planning for autonomous vehicles.

Autonomous vehicles rely on regular data updates for navigation, e.g., detailed road maps, as well as updates in unexpected traffic situations, such as congestion, rain, or black ice. As such, autonomous driving requires fast and reliable communication networks that can deliver such regular data updates. 5G networks provide the benefit of network slicing in which the wireless network is subdivided into virtual network levels. One network level can then be used only for automated driving, as an example. This ensures that safety-relevant notifications to self-driving cars will not end up in a traffic jam on the data highway and will be given priority over other infotainment services used in parallel. Another benefit is the data processing and storage in data centers that are in close proximity to the transport routes. Such “edge” data centers ensure that data can be processed faster in the network. The virtual network levels and short transmission paths can provide high quality of service (QOS) including low latencies and high bandwidths.

While 5G networks can enable fully automated driving, a problem occurs when due to communication congestion, malfunctions or other reasons, wireless network services along a route cannot provide sufficient QoS to ensure full driving information and safety.

It would therefore be advantageous to have some preparation for such eventualities so as to ensure safety and provide sufficient quality of service for automated driving.

The present disclosure provides for using one or more intelligent platforms to determine network traffic conditions at respective network sites and thereby map optimized routes for autonomous vehicles that ensure adequate network coverage and/or quality assurance (QA) for facilitating the operations of the autonomous vehicles along the optimized routes to their respective destinations.

In particular, the disclosure relates to a system and method that utilizes a machine learning (ML) process trained on historical network traffic at respective network sites to process current traffic data to ascertain network traffic conditions along one or more potential routes to a destination for an autonomous vehicle. Accordingly, taking into account predicted use of a particular site at different times of the day based on when an autonomous vehicle is expected to be proximate the site along the one or more potential routes, an optimized route that ensures adequate network coverage and/or QA is determined and assigned to the autonomous vehicle.

In a general aspect, a method is provided. The method includes receiving, at an application function of a core network from a user equipment (UE) device communicatively coupled to an automated vehicle, a navigation destination for the automated vehicle. The method further includes conducting, at the application function, a network health check and route validation process to determine an optimized route from a current location of the automated vehicle to the received navigation destination based on communication network traffic data. The method further includes assigning, at the application function, an optimized route with network quality (NQ) assurance. The method further includes transmitting, to the UE device, an instruction message containing the assigned optimized route.

Implementations of the method can include one or more of the following features:

The method can acknowledge, at the UE device connected to a radio access network apparatus, the assigned optimized route. The method can further include executing, at the radio access network apparatus, NQ assurance and determining network performance associated with the assigned optimized route. The method can further include, upon completion of the NQ assurance and network performance determination, providing instructions at the UE device to cause the automated vehicle to execute the assigned optimized route.

The health check and route validation process can include obtaining, using an rApp function at one or more intelligent platforms, current network traffic data from one or more radio access networks associated with one or more potential routes from the current location of the automated vehicle to the received navigation destination. The health check and route validation process can further include processing the obtained current network traffic data.

The processing can be executed using one or more machine learning (ML) models trained on historical network congestion data previously obtained from the one or more radio access networks.

The historical network congestion data can be obtained using the rApp function to at least periodically stream traffic data from the one or more radio access networks and to store the streamed traffic data in one or more databases.

The one or more machine learning models can be trained using an unsupervised training process for identifying one or more patterns of network traffic congestion at the one or more radio access networks.

The current network traffic data can be processed to predict network congestion at the one or more radio access networks associated with the one or more potential routes.

The optimized route can be assigned according to the one or more radio access networks associated with the one or more potential routes meeting one or more network requirements as determined by the processing, and the one or more network requirements can be associated with the NQ assurance.

In another general aspect, a system is provided. The system includes an interface adapted to communicate with one or more user equipment (UE) devices, a processor and a non-transitory computer-readable memory operatively connected to the processor and having stored thereon machine-readable instructions that cause, when executed, the processor to perform operations. The operations include a step to receive from a user equipment (UE) device communicatively coupled to an automated vehicle, a navigation destination for the automated vehicle. The operations further include a step to conduct a network health check and route validation process to determine an optimized route from a current location of the automated vehicle to the received navigation destination based on communication network traffic data. The operations further include a step to assign an optimized route with network quality (NQ) assurance to the automated vehicle. The operations further include a step to transmit, to the UE device, an instruction message containing the assigned optimized route.

Implementations of the system can include one or more of the following features:

The network health check and route validation process can include machine-readable instructions that cause, when executed, the processor to further obtain, using an rApp function at one or more intelligent platforms, current network traffic data obtained from one or more radio access networks associated with one or more potential routes from the current location of the automated vehicle to the received navigation destination. The network health check and route validation process can include machine-readable instructions that cause, when executed, the processor to further process the obtained current network traffic data.

The obtained current network traffic data can be processed using one or more machine learning models trained on historical network congestion data previously obtained from the one or more radio access networks.

The historical network congestion data can be obtained using the rApp function to at least periodically stream traffic data from the one or more radio access networks and to store the streamed traffic data in one or more databases.

The one or more machine learning models can be trained using an unsupervised training process for identifying one or more patterns of network traffic congestion at the one or more radio access networks.

The current network traffic data can be processed at the processor to predict network congestion at the one or more radio access networks associated with the one or more potential routes.

The optimized route can be assigned according to the one or more radio access networks associated with the one or more potential routes meeting one or more network requirements as determined by the processing, and the one or more network requirements can be associated with the NQ assurance.

In another general aspect, a method is provided for transmitting, from a user equipment (UE) device to a radio access network, a request for a route to a navigation destination, said UE device being communicatively coupled to an automated vehicle. The method further includes receiving from the radio access network an assigned optimized route with network quality (NQ) assurance at the UE device from the radio access network. The method further includes providing instructions, at the UE device for causing the automated vehicle to navigate according to the assigned optimized route upon receiving a confirmation from the radio access network that a NQ assurance has been completed.

Implementations of the method can include one or more of the following features:

The method can further include, after the receiving the optimized route at the UE device connected to the radio access network apparatus, acknowledging the assigned optimized route. The method can further include performing the instructing upon receiving a NQ assurance and network performance determination from the radio access network apparatus.

The optimized route can be assigned according to a network health check and route validation process that comprises processing current network traffic data obtained from one or more radio access networks associated with one or more potential routes from a current location of the automated vehicle to the navigation destination.

The current network traffic data can be processed to predict network congestion at the one or more radio access networks associated with the one or more potential routes.

The optimized route can be assigned according to the one or more radio access networks associated with the one or more potential routes meeting one or more network requirements as determined by the processing, and the one or more network requirements can be associated with the NQ assurance.

These and other aspects, features, and advantages can be appreciated from the following description of certain embodiments and the accompanying drawing figures and claims.

is a schematic diagram that depicts an exemplary scenario illustrating a system and method of route planning for an autonomous vehicle according to one or more embodiments of the present disclosure. In the example depicted in, an autonomous vehicleis shown positioned at an initial Point A. The autonomous vehicleis instructed (e.g., by a passenger) to travel from Point A to Point B. There are alternative routes available to the vehicle, for example, routesandillustrated in. Along a first routefrom Point A to Point B, there are three (3) radio access network (RAN) solution sites with base transceiver stations (BTSs), such as evolved note Bs (eNBs) and/or next generation network nodes (gNodeBs or gNBs) (see, e.g., gNBin), that are in range of the vehicle routes for providing network coverage and services to autonomous vehicle, namely, Sitewith RAN solution (or simply RAN)-, Sitewith RAN-, and Sitewith RAN-. Along a second route, there are two (2) RAN solution sites, Sitewith RAN-and Sitewith RAN-. As illustrated in, Siteand RAN-are available to provide network services along portions of both routesand. In one or more exemplary implementations of the present disclosure, each of the RANscomprises one or more gNBs (in) and is communicatively coupled to a core network (CN)to form a cellular network architecture that conforms to a current generation radio communication standard protocol, such as the 5G network specifications (or standards) described in the 3Generation Partnership Project (3GPP) Technical Specification (TS)., which is incorporated herein by reference as if set forth in its entirety.

The RANsinclude air interface and transceiver portions (e.g., cell towers and associated components) as well as software for implementing central units (CUs) (in) and distributed units (DUs) (in) of, e.g., a 5G network.illustrate an Open Radio Access Network (Open RAN or O-RAN) domain architecture for intelligent platformsand RAN(s)that conforms to the O-RAN specifications set forth by the O-RAN Alliance, which are incorporated herein by reference in their entirely. According to one of more exemplary implementations, CUs provide support for higher layers of the protocol stack, such as service data adaptation protocol (SDAP) and radio resource control (RRC), while DUs provide support for lower layers of the protocol stack, such as media access control (MAC) and the physical layer. In one or more exemplary embodiments, each RANincludes a single CU, while each CU controls multiple DUs (e.g.,-DUs, m>=2 in). Each DU can support one or more cells, or corresponding one or more radio units (RUs), so that a single RAN solutioncan control numerous individual cells. In certain embodiments, one or more RANscan include a singular DU (e.g.,-, m>=1).

In certain embodiments, one or more of RANs, CN, and intelligent platformscan incorporate network elements that conform to standards or protocols other than those of a 5G network for operating in other network environments, such as 2G, 2.5G, 3G, 4G, Wi-Fi, WiMAX, Citizens Broadband Radio Service (CBRS) to name a few. One of ordinary skill in the art will appreciate that while the network elements are described with respect to a 5G network, other network standards are applicable to the present disclosure without departing from its spirit and scope. For example, the present disclosure contemplates a current generation radio communication standard protocol that advances beyond the 5G network standards, for example, to a 6G set of standards and so on.

As described herein, a network environment or solution (sometimes referred to herein simply as a network or an environment) refers to multiple apparatuses, modules, elements, and/or functions that incorporate hardware and/or software and operate to form one or more CNs and one or more Ans that enable wireless communication for a UE device. As described herein, a UE device includes any subscriber device that is communicatively coupled to or integrated with a vehicle control computation system, for example, a cellular telephone device, a mobile computing device, a personal computing device, an onboard vehicle control computation system with a user interface, to name a few.

Referring again to the route map shown in, while RAN-of Siteis functioning, RAN-of Sitecan, for example, suffer from network congestion, as denoted by an alert symbol in. Accordingly, even though the second routeis shorter in terms of distance and would be preferable, for example, based on travel time, it can be problematic because wireless network coverage for vehiclecould be lost or suffer degradation along this route due to the congestion at RAN-. Conversely, the first routeis longer but there is a far greater likelihood of sufficient network QoS to support autonomous driving along this second route if it is determined that neither RAN-nor RAN-suffers from network congestion.

As illustrated in, each of the RAN solutionsis communicatively coupled to a rAPPthat is executed at intelligent platforms—for example, in an O-RAN domain. In one or more exemplary implementations, rAPPis implemented by a RAN Intelligence Controller (RIC), which is implemented at intelligent platforms. According to one or more exemplary implementations, RICconforms to a non-real time service management and orchestration (SMO) framework. Correspondingly, RAN(s), which incorporates CU(s) (in) and DU(s) (in), communicate with RIC/SMOand rAppvia a near-real time (near-RT) RIC (in) and an xApp (in). Thus, near-RT RIC (in) and xApp (in) serve as communication interfaces between RIC(and rApp) and RAN(s). In certain embodiments, they can also be used to implement one or more features described herein. One of ordinary skill in the art will appreciate that features of the present disclosure can be applied to alternative RAN management environments.

Each RAN solution(s)continuously streams telemetry data to rAPP. The telemetry data streamed by RAN solution(s)includes traffic utilization and loading information, for example, from every Site-, from which congestion conditions can be determined. In certain embodiments, the streamed telemetry data can be obtained by rAPPvia CUs (in) and/or DUs (in) of respective RAN(s).

The rAPPstores the obtained telemetry information (or “historical network congestion data”) in one or more historical databases of traffic utilization and loading information, referred to for convenience as “historical network congestion databases.” As illustrated in, CNincorporates an intelligent layer, which communicates with autonomous vehiclevia RANsbased on the traffic data obtained by rApp.

According to one or more exemplary implementations, one or more machine learning (ML) models or algorithmsexecuted at RICare trained using the traffic utilization and loading information derived from the RANsvia rApp. In some embodiments, the ML algorithms are unsupervised.is a schematic diagram showing an overview of an unsupervised ML training processaccording to one or more exemplary implementations of the present disclosure. As shown, raw datacomprising the historical network congestion data, for example, obtained from RAN(s)and stored in one or more databases via rApp, is input to an unsupervised ML model. The raw datais unlabeled, meaning that it has not received a classification from a human user or another ML classifier. The ML modelexecutes a training process based on this raw datato learn patternsin the data. Such patternscan include, but are not limited to, patterns of congestion at one or more network locations—for example, at Sites-—over time, variations in patterns of network congestion by geography, variations in network congestion along different traffic routes—for example, routesand, to name a few. Once such patternshave been recognized, the machine learning algorithmcan be used to process current data to yield predictions about the current state of network congestion in a network, for example, RAN(s).

is similar tobut instead of inputting historical data, current network congestion data, for example, obtained from RAN(s)via rAPP, is input to a trained ML model, and the outputfrom the model provides predictions and analysis of the current state of congestion of a network, for example, RAN(s). Thus, ML modelsandshown inare applicable for ML model(s)shown in.

A number of unsupervised learning techniques can be used by machine learning model(s)trained and executed at RIC, for example, modelsand. Unsupervised machine learning technique can generally be associated with one of three categories: clustering, dimension reduction, and data mining. Clustering techniques are used to sort the raw data into clusters based on mutual similarities. Exemplary clustering techniques that can be used by model(s)of rAPP, for example, modelsand, include, but are not limited to, K-means, Hierarchical Clustering and Fuzzy C-means. Dimensionality reduction techniques reduce the dimensions of the raw data to reduce computational complexity and throughput. As an example, 4-dimensional arrays of data can be reduced to 3-dimensional arrays using techniques such as, but not limited to, principal component analysis (PCA) and linear discriminant reduction (LCR). Data mining techniques are used to determine relationships between variables in the raw data. For example, data mining can discover association rules. In the present context, one example is using data to associate certain times of day/locations pairs with elevated levels of congestion. Some of the more prevalently used data mining algorithms include Naïve Bayes, K-means, and Apriori. However, other data mining algorithms can be used.

Unsupervised machine learning techniques are particularly suited for anomaly detection, which can include unusually high or low levels of traffic and congestion or network loading. Furthermore, the unsupervised machine learning system can provide for wireless customer segmentation, by sorting customers according to travel and wireless resource use behavior, among other applications.

is a schematic diagram illustrating a network environmentaccording to one or more exemplary implementations of the present disclosure. Network environmentincludes one or more automated vehicles-. . .-(n>=1), which can be partially or fully autonomous vehicles. For example, automated vehicle(s)can include automated cars, buses, trucks, trams, utility vehicles, maintenance vehicles, construction vehicles, public transit vehicles, to name a few. In general, automated vehicle(s)include various sensors and automated controls (including automated driving controls) by which a computational system can direct the travel (e.g., direction, speed, path, etc.) of the vehicle fully or partly without human intervention. The vehicle control computation system for each automated vehicleincludes one or more onboard processors, an onboard vehicle controller, and one or more position sensors, which are illustrated for vehicle-inas a representative example for vehicle(s).

The one or more on-board computational processors (OBCPs)can include one or more processor devices (/in) and one or more memory devices (/in), such as non-transient processor-readable memory devices. In some embodiments, each OBCPincludes a network interface (e.g., including any suitable antennas, transceivers, etc.) by which the automated vehiclehaving the OBCPcan communicate with CNvia a gNBand corresponding RAN. Processor(s)and controllerperform navigational tasks for their vehiclethat involve processing data received from one or more position sensors. In exemplary embodiments, sensor(s)can include global positioning satellite (GPS) devices, radar sensors, infrared (IR) sensors, light detection and ranging (LiDAR) sensors, cameras, proximity sensors, to name a few. Accordingly, the vehicle control computation system performs tasks that involve controlling, and processing feedback relating to braking, accelerating, steering, and other driving functions. Additional navigational tasks can involve detecting and responding to traffic rules and conditions, traffic events, such as traffic accidents, roadway damage, roadway obstructions, slow-downs, and the like.

Automated vehicle(s)are in wireless communication with CNvia respective one or more gNBs-. . .-(o>=1) and corresponding one or more RANs-. . .-(p>=1). RAN-is illustrated as a representative example of RAN(s)-. . .-with CUand one or more DUs-. . .-(m>=2) in communication with rAppexecuted at intelligent platforms. One of ordinary skill in the art will appreciate that network environmentcan include additional RANs, such as RANs-. . .-illustrated in, corresponding to additional gNBs, and additional automated vehiclesin communication therewith. In some embodiments, RAN(s)can be coupled to one or more BTSs operating in an environment other than a 5G environment, for example, eNBs or the like. Some or all automated vehiclescan include antennas and transceivers through which they communicate with nearby cellular towers (or RUs) of gNBs, with one or more satellites, with one or more nearby relay stations, and/or with any other infrastructure that can send and receive wireless communications. In certain embodiments, vehiclescan communicate with CNvia another CN and/or associated RAN(s), gNB(s), BTS(s), eNB(s), or the like.

According to one or more exemplary implementations, network traffic data related to RAN(s)is obtained via intelligent platforms, where rAppof non-RT RIC/SMOobtains network traffic data from RAN(s)via near-RT RICand/or xApp. Accordingly, historical network congestion data from RAN(s)is at least periodically obtained and stored in one or more databases (e.g., maintained at storage devicein), which can be maintained at one or more computing apparatuses (e.g.,in) comprised in or communicatively coupled to intelligent platforms. The stored historical network congestion data is used (e.g., as raw datain) to train one or more ML models, for example, model, comprised in model(s)to obtain one or more trained models, for example, model, also comprised in model(s). The trained model(s)can then be applied to analyze subsequently obtained network traffic data from one or more RANsin correspondence with one or more vehiclesto determine one or more suitable routes from the current location(s) of the vehicle(s)(e.g., Point A in) to the navigation destination(s) (e.g., Point B in) that would ensure QoS and/or network coverage for providing navigational instructions for automated operations of vehicle(s).

also illustrates domain network elements of CNfor communicating with vehicle(s)in a manner of description associated with service-based system architectures with network elements, or network functions, interconnected via a common bus that operates in compliance with the 5G network standards. Accordingly, network function messages in CNare communicated among the network elements shown in the CNdomain to affect higher layer services associated with vehicle(s), for example, navigational instructions for automated operations of vehicle(s). The lines depicted inconnecting the network elements in the CNdomain indicate possible direct communications among any two or more of the network elements. Certain processes of the present disclosure can be executed over point-to-point interfaces among the depicted network elements as can be appreciated by one of ordinary skill in the art. As one of ordinary skill in the art will also appreciate, the network elements in the CNdomain illustrated incan be embodied as multiple respective instances serving respective ones of multiple subscribers and/or vehicles.

As illustrated in, the CNdomain embodies a packet core network that comprises an access and mobility management function (AMF), which supports encrypted signaling associated with a user equipment (UE) device, for example, vehicle(s). CNfurther includes a session management function and packet data network gateway-control module (SMF/PGW-C)that manages a session of vehicle(s)and a user plane function and packet data network gateway-user plane module (UPF/PGW-U)that processes and forwards user data, for example, between vehicle(s)and an external data network (not shown), such as the Internet. According to one or more exemplary implementations, AMFforwards messages related to session management between vehicle(s)and SMF/PGW-C. SMF/PGW-Cmanages UE device sessions and, in doing so, interacts with other network functions, including asserting functional control of UPF/PGW-U, for example, for traffic and/or quality of service (QOS) related features. In certain embodiments, AMFcan comprise a mobility management entity (MME) (not shown) for interoperability with 4G networks and, for example, long term evolution (LTE) RANs, which can be embodied by one or more of RANs.

The CNdomain further comprises a network exposure function (NEF)that supports interactions by network functions in the CNdomain with applications that are executed to provide service features to a subscriber (e.g., vehicle(s)) and one or more application functions (AF). NEFprovides communications to and from applications (or AF(s)), including providing for applications to trigger devices—for example, vehicle(s), to execute actions. In accordance with one or more exemplary implementations of the present disclosure, NEFmediates communications between vehicle(s)and one or more Afs, such as one or more application features of intelligent layer, via AMF. Additionally, NEFprovides for data provisioning from AF(s)to AMFfor tuning CNsettings, facilitating state changes for UE devices (e.g., vehicle(s)), and/or optimizing network signaling capacity. NEFfurther provides certain policy (e.g., QoS) and charging controls to AF(s). In certain embodiments, CNcan incorporate direct proprietary connections between one or more of AF(s)and the other network elements, for example, AMF.

Intelligence layeroperates at least in part on an application layer in connection with executing one or more application features for CNand/or vehicle(s). In certain embodiments, intelligence layercan be a separate network element in the CNdomain. According to one or more exemplary implementations, intelligence layerembodies one or more application functions in AFthat provide analytical feedback based on historical network congestion data and/or modeltrained on such data. The network traffic data, or historical network congestion data, accumulated by rAppfor training model/includes but is not limited to RAN network data, performance data, fault management data, network alarms, and network logs. Thus, network coverage and traffic information can be derived for roadway routes and respective locations along such routes based on the historical network congestion data. In certain embodiments, AF(s)can further comprise one or more Application Programming Interfaces (APIs) (not shown) for providing a direct interface to one or more application features.